27 KiB

Raw Blame History

Incompatible changes in ArangoDB 3.4

It is recommended to check the following list of incompatible changes before upgrading to ArangoDB 3.4, and adjust any client programs if necessary.

The following incompatible changes have been made in ArangoDB 3.4:

Supported platforms

The minimum supported architecture for the official release packages of ArangoDB is now the Nehalem architecture.

All release packages are built with compiler optimizations that require at least this architecture. The following CPU features are required for running an official release package (note: these are all included in the Nehalem architecture and upwards):

SSE2
SSE3
SSE4.1
SSE4.2

In case the target platform does not conform to these requirements, ArangoDB may not work correctly.

The compiled-in architecture optimizations can be retrieved on most platforms by invoking the arangod binary with the --version option. The optimization switches will then show up in the output in the line starting with optimization-flags, e.g.

$ arangod --version
...
optimization-flags: -march=nehalem -msse2 -msse3 -mssse3 -msse4.1 -msse4.2 -mno-sse4a -mno-avx -mno-fma -mno-bmi2 -mno-avx2 -mno-xop -mno-fma4 -mno-avx512f -mno-avx512vl -mno-avx512pf -mno-avx512er -mno-avx512cd -mno-avx512dq -mno-avx512bw -mno-avx512ifma -mno-avx512vbmi
platform: linux

Note that to get even more target-specific optimizations, it is possible for end users to compile ArangoDB on their own with compiler optimizations tailored to the target environment.

Storage engine

In ArangoDB 3.4, the default storage engine for new installations is the RocksDB engine. This differs to previous versions (3.2 and 3.3), in which the default storage engine was the MMFiles engine.

The MMFiles engine can still be explicitly selected as the storage engine for all new installations. It's only that the "auto" setting for selecting the storage engine will now use the RocksDB engine instead of MMFiles engine.

In the following scenarios, the effectively selected storage engine for new installations will be RocksDB:

--server.storage-engine rocksdb
--server.storage-engine auto
--server.storage-engine option not specified

The MMFiles storage engine will be selected for new installations only when explicitly selected:

--server.storage-engine mmfiles

To make users aware of that the RocksDB storage engine was chosen automatically due to an explicit other storage engine selection, 3.4 will come up with the following startup warning:

using default storage engine 'rocksdb', as no storage engine was explicitly selected via the `--server.storage-engine` option.
please note that default storage engine has changed from 'mmfiles' to 'rocksdb' in ArangoDB 3.4

On upgrade, any existing ArangoDB installation will keep its previously selected storage engine. The change of the default storage engine in 3.4 is thus only relevant for new ArangoDB installations and/or existing cluster setups for which new server nodes get added later. All server nodes in a cluster setup should use the same storage engine to work reliably. Using different storage engines in a cluster is unsupported.

To validate that the different nodes in a cluster deployment use the same storage engine throughout the entire cluster, there is now a startup check performed by each coordinator. Each coordinator will contact all DB servers and check if the same engine on the DB server is the same as its local storage engine. In case there is any discrepancy, the coordinator will abort its startup.

Geo indexes

The on-disk storage format for indexes of type geo has changed for the RocksDB storage engine. This also affects geo1 and geo2 indexes.

This requires users to start the arangod process with the --database.auto-upgrade true option to allow ArangoDB recreating these indexes using the new on-disk format.

The on-disk format for geo indexes is incompatible with the on-disk format used in 3.3 and 3.2, so an in-place downgrade from 3.4 to 3.3 is not supported.
Geo indexes will now be reported no longer as geo1 or geo2 but as type geo. The two previously known geo index types (geo1and geo2) are deprecated. APIs for creating indexes (ArangoCollection.ensureIndex) will continue to support geo1and geo2.

RocksDB engine data storage format

Installations that start using ArangoDB 3.4 will use an optimized on-disk format for storing documents using the RocksDB storage engine. The RocksDB engine will also a new table format version that was added in a recent version of the RocksDB library and that is not available in ArangoDB versions before 3.4.

This format cannot be used with ArangoDB 3.3 or before, meaning it is not possible to perform an in-place downgrade from a fresh 3.4 install to 3.3 or earlier when using the RocksDB engine. For more information on how to downgrade, please refer to the Downgrading chapter.

Installations that were originally set up with older versions of ArangoDB (e.g. 3.2 or 3.3) will continue to use the existing on-disk format for the RocksDB engine even with ArangoDB 3.4 (unless you install a fresh 3.4 package and restore a backup of your data on this fresh installation).

In order to use the new binary format with existing data, it is required to create a logical dump of the database data, shut down the server, erase the database directory and restore the data from the logical dump. To minimize downtime you can alternatively run a second arangod instance in your system, that replicates the original data; once the replication has reached completion, you can switch the instances.

RocksDB intermediate commits

Intermediate commits in the rocksdb engine are now only enabled in standalone AQL queries (not within a JS transaction), standalone truncate as well as for the "import" API.

The options intermediateCommitCount and intermediateCommitSize will have no affect anymore on transactions started via /_api/transaction, or db._executeTransaction().

Threading and request handling

The processing of incoming requests and the execution of requests by server threads has changed in 3.4.

Previous ArangoDB versions had a hard-coded implicit lower bound of 64 running threads, and up to which they would increase the number of running server threads. That value could be increased further by adjusting the option --server.maximal-threads. The configuration option --server.threads existed, but did not effectively set or limit the number of running threads.

In ArangoDB 3.4, the number of threads ArangoDB uses for request handling can now be strictly bounded by configuration options.

The number of server threads is now configured by the following startup options:

--server.minimal-threads: determines the minimum number of request processing threads the server will start and always keep around
--server.maximal-threads: determines the maximum number of request processing threads the server will start for request handling. If that number of threads is already running, arangod will not start further threads for request handling

The actual number of request processing threads is adjusted dynamically at runtime and will float between --server.minimal-threads and --server.maximal-threads.

HTTP REST API

The following incompatible changes were made in context of ArangoDB's HTTP REST APIs:

The following, partly undocumented internal REST APIs have been removed in ArangoDB 3.4:
- GET /_admin/test
- GET /_admin/clusterCheckPort
- GET /_admin/cluster-test
- GET /_admin/routing/routes
- GET /_admin/statistics/short
- GET /_admin/statistics/long
- GET /_admin/auth/reload
GET /_api/index will now return type geo for geo indexes, not type geo1 or geo2 as previous versions did.

For geo indexes, the index API will not return the attributes constraint and ignoreNull anymore. These attributes were initially deprecated in ArangoDB 2.5
GET /_api/aqlfunction was migrated to match the general structure of ArangoDB replies. It now returns an object with a "result" attribute that contains the list of available AQL user functions:
```
{
  "code": 200,
  "error": false,
  "result": [
    {
      "name": "UnitTests::mytest1",
      "code": "function () { return 1; }",
      "isDeterministic": false
    }
  ]
}
```
In previous versions, this REST API returned only the list of available AQL user functions on the top level of the response. Each AQL user function description now also contains the 'isDeterministic' attribute.
if authentication is turned on, requests to databases by users with insufficient access rights will be answered with HTTP 401 (forbidden) instead of HTTP 404 (not found).

The following APIs have been added or augmented:

POST /_api/document/{collection} now supports repsert (replace-insert).

This can be achieved by using the URL parameter overwrite=true. When set to true, insertion will not fail in case of a primary key conflict, but turn into a replace operation.

When an insert turns into a replace, the previous version of the document can be retrieved by passing the URL parameter returnOld=true
POST /_api/aqlfunction now includes an "isNewlyCreated" attribute that indicates if a new function was created or if an existing one was replaced (in addition to the "code" attribute, which remains 200 for replacement and 201 for creation):
```
{
  "code": "201",
  "error": false,
  "isNewlyCreated": true
}
```

DELETE /_api/aqlfunction now returns the number of deleted functions:

{
  "code": 200,
  "error": false,
  "deletedCount": 10
}

GET /_admin/status now returns the attribute operationMode in addition to mode. The attribute writeOpsEnabled is now also represented by the new attribute readOnly, which is has an inverted value compared to the original attribute. The old attributes are deprecated in favor of the new ones.
POST /_api/collection now will process the optional shardingStrategy attribute in the response body in cluster mode.

This attribute specifies the name of the sharding strategy to use for the collection. Since ArangoDB 3.4 there are different sharding strategies to select from when creating a new collection. The selected shardingStrategy value will remain fixed for the collection and cannot be changed afterwards. This is important to make the collection keep its sharding settings and always find documents already distributed to shards using the same initial sharding algorithm.

The available sharding strategies are:
- community-compat: default sharding used by ArangoDB community versions before ArangoDB 3.4
- enterprise-compat: default sharding used by ArangoDB enterprise versions before ArangoDB 3.4
- enterprise-smart-edge-compat: default sharding used by smart edge collections in ArangoDB enterprise versions before ArangoDB 3.4
- hash: default sharding used by ArangoDB 3.4 for new collections (excluding smart edge collections)
- enterprise-hash-smart-edge: default sharding used by ArangoDB 3.4 for new smart edge collections
If no sharding strategy is specified, the default will be hash for all collections, and enterprise-hash-smart-edge for all smart edge collections (requires the Enterprise Edition of ArangoDB). Manually overriding the sharding strategy does not yet provide a benefit, but it may later in case other sharding strategies are added.

In single-server mode, the shardingStrategy attribute is meaningless and will be ignored.
APIs for view management have been added at endpoint /_api/view.
The REST APIs for modifying graphs at endpoint /_api/gharial now support returning the old revision of vertices / edges after modifying them. The APIs also supports returning the just-inserted vertex / edge. This is in line with the already existing single-document functionality provided at endpoint /_api/document.

The old/new revisions can be accessed by passing the URL parameters returnOld and returnNew to the following endpoints:
- /_api/gharial//vertex/
- /_api/gharial//edge/
The exception from this is that the HTTP DELETE verb for these APIs does not support returnOld because that would make the existing API incompatible.

AQL

the AQL functions CALL and APPLY may now throw the errors 1540 (ERROR_QUERY_FUNCTION_NAME_UNKNOWN) and 1541 (ERROR_QUERY_FUNCTION_ARGUMENT_TYPE_MISMATCH) instead of error 1582 (ERROR_QUERY_FUNCTION_NOT_FOUND) in some situations.
the existing "fulltext-index-optimizer" optimizer rule has been removed because its duty is now handled by the new "replace-function-with-index" rule.
the behavior of the fullCount option for AQL queries has changed so that it will only take into account LIMIT statements on the top level of the query.

LIMIT statements in subqueries will not have any effect on the fullCount results any more.
the AQL functions NEAR, WITHIN, WITHIN_RECTANGLE and FULLTEXT do not support accessing collections dynamically anymore.

The name of the underlying collection and the name of the index attribute to be used have to specified using either collection name identifiers, string literals or bind parameters, but must not be specified using query variables.

For example, the following AQL queries are ok:
```
FOR doc IN NEAR(myCollection, 2.5, 3) RETURN doc
FOR doc IN NEAR(@@collection, 2.5, 3) RETURN doc
FOR doc IN FULLTEXT("myCollection", "body", "foxx") RETURN doc
FOR doc IN FULLTEXT(@@collection, @attribute, "foxx") RETURN doc
```
Contrary, the following queries will fail to execute with 3.4 because of dynamic collection/attribute names used in them:
```
FOR name IN ["col1", "col2"] FOR doc IN NEAR(name, 2.5, 3) RETURN doc

FOR doc IN collection 
  FOR match IN FULLTEXT(PARSE_IDENTIFIER(doc).collection, PARSE_IDENTIFIER(doc).key, "foxx") RETURN doc
```
the AQL warning 1577 ("collection used in expression") will not occur anymore

It was used in previous versions of ArangoDB when the name of a collection was used in an expression in an AQL query, e.g.
```
RETURN c1 + c2
```
Due to internal changes in AQL this is not detected anymore in 3.4, so this particular warning will not be raised.

Additionally, using collections in arbitrary AQL expressions as above is unsupported in a mixed cluster that is running a 3.3 coordinator and 3.4 DB server(s). The DB server(s) running 3.4 will in this case not be able to use a collection in an arbitrary expression, and instead throw an error.
the undocumented built-in visitor functions for AQL traversals have been removed, as they were based on JavaScript implementations:
- HASATTRIBUTESVISITOR
- PROJECTINGVISITOR
- IDVISITOR
- KEYVISITOR
- COUNTINGVISITOR
Using any of these functions from inside AQL will now produce an error.

in previous versions, the AQL optimizer used two different ways of converting strings into numbers. The two different ways have been unified into a single way that behaves like the TO_NUMBER AQL function, which is also the documented behavior.

The change affects arithmetic operations with strings that contain numbers and other trailing characters, e.g.

expression         3.3 result          3.4 result       TO_NUMBER()
0 + "1a"           0 + 1 = 1           0 + 0 = 0        TO_NUMBER("1a") = 0
0 + "1 "           0 + 1 = 1           0 + 1 = 1        TO_NUMBER("1 ") = 1
0 + " 1"           0 + 1 = 1           0 + 1 = 1        TO_NUMBER(" 1") = 1
0 + "a1"           0 + 0 = 0           0 + 0 = 0        TO_NUMBER("a1") = 0

the AQL function DATE_NOW is now marked as deterministic internally, meaning that the optimizer may evaluate the function at query compile time and not at query runtime. This will mean that calling the function repeatedly inside the same query will now always produce the same result, whereas in previous versions of ArangoDB the function may have generated different results.

Each AQL query that is run will still evalute the result value of the DATE_NOW function independently, but only once at the beginning of the query. This is most often what is desired anyway, but the change makes DATE_NOW useless to measure time differences inside a single query.
the internal AQL function PASSTHRU (which simply returns its call argument) has been changed from being non-deterministic to being deterministic, provided its call argument is also deterministic. This change should not affect end users, as PASSTHRU is intended to be used for internal testing only. Should end users use this AQL function in any query and need a wrapper to make query parts non-deterministic, the NOOPT AQL function can stand in as a non-deterministic variant of PASSTHRU

Usage of V8

The internal usage of the V8 JavaScript for non-user actions has been reduced in ArangoDB. Several APIs have been rewritten to not depend on V8 and thus do not require using a V8 context for execution.

Compared to ArangoDB 3.3, the following parts of ArangoDB can now be used without requiring V8 contexts:

all of AQL (with the exception of user-defined functions)
the graph modification APIs at endpoint /_api/gharial
background server statistics gathering

Reduced usage of V8 by ArangoDB may allow end users to lower the configured numbers of V8 contexts to start. In terms of configuration options, these are:

--javascript.v8-contexts: the maximum number of V8 contexts to create
--javascript.v8-contexts-minimum: the minimum number of V8 contexts to create at server start and to keep around

The default values for these startup options have not been changed in ArangoDB 3.4, but depending on the actual workload, 3.4 ArangoDB instances may need less V8 contexts than 3.3.

Startup option changes

For arangod, the following startup options have changed:

the number of server threads is now configured by the following startup options:
- --server.minimal-threads: determines the minimum number of request processing threads the server will start
- --server.maximal-threads: determines the maximum number of request processing threads the server will start
The actual number of request processing threads is adjusted dynamically at runtime and will float between --server.minimal-threads and --server.maximal-threads.
the default value for the existing startup option --javascript.gc-interval has been increased from every 1000 to every 2000 requests, and the default value for the option --javascript.gc-frequency has been increased from 30 to 60 seconds.

This will make the V8 garbage collection run less often by default than in previous versions, reducing CPU load a bit and leaving more V8 contexts available on average.
the startup option --cluster.my-local-info has been removed in favor of persisted server UUIDs.

The option --cluster.my-local-info was deprecated since ArangoDB 3.3.
the startup option --database.check-30-revisions was removed. It was used for checking the revision ids of documents for having been created with ArangoDB 3.0, which required a dump & restore migration of the data to 3.1.

As direct upgrades from ArangoDB 3.0 to 3.4 or from 3.1 to 3.4 are not supported, this option has been removed in 3.4.
the startup option --server.session-timeout has been obsoleted. Setting this option will not have any effect.
the option --replication.automatic-failover was renamed to --replication.active-failover

Using the old option name will still work in ArangoDB 3.4, but support for the old option name will be removed in future versions of ArangoDB.
the option --rocksdb.block-align-data-blocks has been added

If set to true, data blocks stored by the RocksDB engine are aligned on lesser of page size and block size, which may waste some memory but may reduce the number of cross-page I/Os operations.

The default value for this option is false.

Permissions

The behavior of permissions for databases and collections changed:

The new fallback rule for databases for which no access level is explicitly specified is now:

Choose the higher access level of:
- A wildcard database grant
- A database grant on the _system database

The new fallback rule for collections for which no access level is explicitly specified is now:

Choose the higher access level of:
- Any wildcard access grant in the same database, or on "/"
- The access level for the current database
- The access level for the _system database

SSLv2

Support for SSLv2 has been removed from arangod and all client tools.

Startup will now be aborted when using SSLv2 for a server endpoint, or when connecting with one of the client tools via an SSLv2 connection.

SSLv2 has been disabled in the OpenSSL library by default in recent versions because of security vulnerabilities inherent in this protocol.

As it is not safe at all to use this protocol, the support for it has also been stopped in ArangoDB. End users that use SSLv2 for connecting to ArangoDB should change the protocol from SSLv2 to TLSv12 if possible, by adjusting the value of the --ssl.protocol startup option.

Mixed-engine clusters

Starting a cluster with coordinators and DB servers using different storage engines is not supported. Doing it anyway will now log an error and abort a coordinator's startup.

Previous versions of ArangoDB did not detect the usage of different storage engines in a cluster, but the runtime behavior of the cluster was undefined.

Client tools

The client tool arangoimp has been renamed to arangoimport for consistency.

Release packages will still install arangoimp as a symlink to arangoimport, so user scripts invoking arangoimp do not need to be changed to work with ArangoDB 3.4.

In the ArangoShell, the undocumented JavaScript module @arangodb/actions has been removed. This module contained the methods printRouting and printFlatRouting, which were used for debugging purposes only.

In the ArangoShell, the undocumented JavaScript functions reloadAuth and routingCache have been removed from the internal module.

Foxx applications

The undocumented JavaScript module @arangodb/database-version has been removed, so it cannot be use from Foxx applications anymore The module only provided the current version of the database, so any client-side invocations can easily be replaced by using the db._version() instead.

Miscellaneous changes

For the MMFiles engine, the compactor thread(s) were renamed from "Compactor" to "MMFilesCompactor".

This change will be visible only on systems which allow assigning names to threads.

Deprecated features

The following features and APIs are deprecated in ArangoDB 3.4, and will be removed in future versions of ArangoDB:

the JavaScript-based traversal REST API at /_api/traversal:

This API has several limitations (including low result set sizes) and has effectively been unmaintained since the introduction of AQL's general TRAVERSAL clause.

It is recommended to migrate client applications that use the REST API at /_api/traversal to use AQL-based traversal queries instead.
the REST API for simple queries at /_api/simple:

The simple queries provided by the /_api/simple endpoint are limited in functionality and will internally resort to AQL queries anyway. It is advised that client applications also use the equivalent AQL queries instead of using the simple query API, because that is more flexible and allows greater control of how the queries are executed.
the REST API for querying endpoints at /_api/endpoints:

The API /_api/endpoint is deprecated since ArangoDB version 3.1. For cluster mode there is /_api/cluster/endpoints to find all current coordinator endpoints.
accessing collections via their numeric IDs instead of their names. This mostly affects the REST APIs at
- /_api/collection/<collection-id>
- /_api/document/<collection-id>
- /_api/simple
Note that in ArangoDB 3.4 it is still possible to access collections via their numeric ID, but the preferred way to access a collections is by its user-defined name.
the REST API for WAL tailing at /_api/replication/logger-follow:

The logger-follow WAL tailing API has several limitations. A better API was introduced at endpoint /_api/wal/tail in ArangoDB 3.3.

Client applications using the old tailing API at /_api/replication/logger-follow should switch to the new API eventually.
the result attributes mode and writeOpsEnabled in the REST API for querying a server's status at /_admin/status:

GET /_admin/status returns the additional attributes operationMode and readOnly now, which should be used in favor of the old attributes.
creating geo indexes via any APIs with one of the types geo1 or geo2:

The two previously known geo index types (geo1and geo2) are deprecated now. Instead, when creating geo indexes, the type geo should be used.

The types geo1 and geo2 will still work in ArangoDB 3.4, but may be removed in future versions.
the legacy mode for Foxx applications from ArangoDB 2.8 or earlier:

The legacy mode is described in more detail in the Foxx manual. To upgrade an existing Foxx application that still uses the legacy mode, please follow the steps described in the manual.
the AQL geo functions NEAR, WITHIN, WITHIN_RECTANGLE and IS_IN_POLYGON:

The special purpose NEAR AQL function can be substituted with the following AQL (provided there is a geo index present on the doc.latitude and doc.longitude attributes) since ArangoDB 3.2:
```
FOR doc in geoSort
  SORT DISTANCE(doc.latitude, doc.longitude, 0, 0)
  LIMIT 5
  RETURN doc
```
WITHIN can be substituted with the following AQL since ArangoDB 3.2:
```
FOR doc in geoFilter
  FILTER DISTANCE(doc.latitude, doc.longitude, 0, 0) < 2000
  RETURN doc
```
Compared to using the special purpose AQL functions this approach has the advantage that it is more composable, and will also honor any LIMIT values used in the AQL query.

In ArangoDB 3.4, NEAR, WITHIN, WITHIN_RECTANGLE and IS_IN_POLYGON will still work and automatically be rewritten by the AQL query optimizer to the above forms. However, AQL queries using the deprecated AQL functions should eventually be adjusted.

27 KiB Raw Blame History